Oh-kyoung Kwon | Searcharxiv

arXiv2026

From Dense Gas Clouds to Supermassive Black Hole Seeds: Hybrid Hydro/Direct $N$-body Simulations of Runaway Collision-driven Intermediate-mass Black Hole Formation

A population of dense stellar systems at high redshift has recently been uncovered by the JWST. To investigate the formation of supermassive black hole (SMBH) seeds in these dense environments without invoking any \textit{ad hoc} seeding mechanisms, we present star cluster-scale simulations performed with an updated version of the hydrodynamics code \texttt{Enzo-Abyss}, which self-consistently integrates the gravity using a direct $N$-body method coupled with stellar evolution. By modeling initially dense, metal-poor gas clouds with varying turbulence, we consistently find the formation of dense clusters resembling early-stage nuclear star clusters (NSCs), as well as the formation of very massive stars (VMSs) ranging from $343\;\mathrm{M_\odot}$ to $5108\;\mathrm{M_\odot}$ via runaway collisions, irrespective of stellar wind feedback strength. Following the direct collapse of these VMSs, the resulting intermediate-mass black holes (IMBHs) grow through Eddington-limited gas accretion and tidal disruption events (TDEs). In our most optimistic model, we find a mass accretion rate of $1.64\times10^{-4}\;\mathrm{M_\odot\;yr^{-1}}$, with TDEs contributing $23\%$ of the total accretion over $\sim10\;\mathrm{Myr}$. Assuming a steady gas supply into the NSC driven by rapid structural assembly in the high-redshift environment, together with a constant TDE rate, we project that an IMBH with an initial mass of $6747\;\mathrm{M_\odot}$ at the center of the NSC can grow to $\sim62000\;\mathrm{M_\odot}$ within $100\;\mathrm{Myr}$ of its formation. Our numerical study, conducted within a single self-consistent framework that incorporates the essential physical processes, suggests that VMSs can form in dense gas clouds, collapse into IMBHs, and subsequently provide viable seeds for the SMBHs observed at high redshift.

astro-ph.GA↗

arXiv2022

Crossover-SGD: A gossip-based communication in distributed deep learning for alleviating large mini-batch problem and enhancing scalability

Distributed deep learning is an effective way to reduce the training time of deep learning for large datasets as well as complex models. However, the limited scalability caused by network overheads makes it difficult to synchronize the parameters of all workers. To resolve this problem, gossip-based methods that demonstrates stable scalability regardless of the number of workers have been proposed. However, to use gossip-based methods in general cases, the validation accuracy for a large mini-batch needs to be verified. To verify this, we first empirically study the characteristics of gossip methods in a large mini-batch problem and observe that the gossip methods preserve higher validation accuracy than AllReduce-SGD(Stochastic Gradient Descent) when the number of batch sizes is increased and the number of workers is fixed. However, the delayed parameter propagation of the gossip-based models decreases validation accuracy in large node scales. To cope with this problem, we propose Crossover-SGD that alleviates the delay propagation of weight parameters via segment-wise communication and load balancing random network topology. We also adapt hierarchical communication to limit the number of workers in gossip-based communication methods. To validate the effectiveness of our proposed method, we conduct empirical experiments and observe that our Crossover-SGD shows higher node scalability than SGP(Stochastic Gradient Push).

cs.LG↗